Research map / Federated learning and privacy / Differential privacy
FIBER: A Differentially Private Optimizer with Filter-Aware Innovation Bias Correction
Duc Dm, Thao Do, Minh Son Hoang, Anh Le Duc Tran, Daeyoung Kim, Huy Nguyen
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| innovation filtering (FiBeR) instead of DiSK gradient-state smoothing (ablation) | DP-AdamW · CIFAR-10 CNN5 eps=0.5 | accuracy (%): 59.70 -> 65.32 | Improved |
| innovation filtering instead of DiSK smoothing (ablation) | DP-AdamW · CIFAR-10 CNN5 eps=2 | accuracy (%): 68.80 -> 71.39 | Improved |
| innovation filtering instead of DiSK smoothing (ablation) | DP-AdamW · CIFAR-10 CNN5 eps=8 | accuracy (%): 74.90 -> 73.19 | Hurt |
| raise innovation gain omega from 0.60 to 0.90 (ablation) | FiBeR · CIFAR-10 CNN5 eps=1 | accuracy (%): 68.86 -> 70.76 | Improved |
| raise innovation gain omega from 0.90 to 0.99 (ablation) | FiBeR · CIFAR-10 CNN5 eps=1 | accuracy (%): 70.76 -> 70.62 | No clear effect |
| add filter-aware second-moment correction (ablation) | FiBeR · CIFAR-10 across privacy budgets | accuracy | Improved |
| add FiBeR on top of DiSK | DiSK · ViT-small CIFAR-100 fine-tuning eps=0.5 | accuracy (%): 83.5 -> 88.6 | Improved |
| add FiBeR on top of DiSK | DiSK · ViT-small CIFAR-100 fine-tuning eps=8 | accuracy (%): 88.5 -> 90.5 | Improved |
| add FiBeR | DiSK · ImageNet-1k ViT-small | accuracy: 39 -> 44 | Improved |
| add FiBeR | DPAdamW · GLUE QNLI eps=1 | accuracy: 86.0 -> 90.5 | Improved |
| add FiBeR vs DiSK | DiSK · GLUE MNLI eps=6.7 | accuracy: 84.8 -> 84.7 | Hurt |
| add FiBeR vs DiSK | DiSK · E2E BLEU eps=3 | BLEU: 68.35 -> 67.57 | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- DiSK — DiSK: Differentially Private Optimizer with Simplified Kalman Filter for Noise Reduction (ICLR 2025) · better on 10 of 11 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- DiSK: Differentially Private Optimizer with Simplified Kalman Filter for Noise Reduction (ICLR 2025 · similarity 0.89)
- DOPPLER: Differentially Private Optimizers with Low-pass Filter for Privacy Noise Reduction (NeurIPS 2024 · similarity 0.88)
- Differentially Private Bias-Term Fine-tuning of Foundation Models (ICML 2024 · similarity 0.86)
- Towards hyperparameter-free optimization with differential privacy (ICLR 2025 · similarity 0.85)
- Privacy without Noisy Gradients: Slicing Mechanism for Generative Model Training (NeurIPS 2024 · similarity 0.85)
- Subsampling is not Magic: Why Large Batch Sizes Work for Differentially Private Stochastic Optimisation (ICML 2024 · similarity 0.85)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
